Summarization of Cricket Videos Using Deep Learning Technique

Abstract

Video Summarization plays a ignificant role in many fields of life and it can be employed to avoid wastage of time and effort in watching long and boring different types of sports videos. In literature, several computer vision-based techniques have been proposed for predicting useful classes from different perspectives, which is quite challenging. The prediction accuracy results of previous techniques were not satisfactory due to the complex nature and lot of redundant events in sports videos. This work focuses on the video summarization of cricket videos due to their overwhelming interest. To make predictions accurately and precisely, a well-organized dataset of four main classes of cricket like, catch, lbw, four, and sixer as the viewer is not interested in unnecessary coverage of the crowd and replays that just waste its time and interest. In the experiments, cricket videos were extracted from different sources, especially YouTube. Subsequently, these videos have been processed and the most useful frames were extracted to run several experiments. The Resnet152 V2 transfer learning model was implemented to carry out the classification and video summarization task. The proposed method performs well and produces more accurate results. The approach of making datasets reduces inter-class similarity problems that occurred in previous methods of video summarization. The proposed work will be helpful in saving the time of viewers by viewing and observing summarized videos instead of long content videos.